Your point about having a version 1 and iterating quickly based on real world input reminds me of my experience as a mechanical design engineer. The most costly mistakes made in that field resulted from too much design and not enough prototype phases. The typical design process back in the 80's was to meticulously draw every aspect of a component, send it to a machine shop for tooling, get manufactured parts and put them on the test stand. Only then would you discover that the gage was too light, the material was the wrong spec or the welds had to be moved.
I introduced the concept of prototype early and often. Actually re-introduced, as this is how most products were engineered in the olden days. Someone would suggest a new gage, weld placement, cross section, and I would don my overalls, head to the shop, cobble up something and demonstrate its feasibility before the draftsmen could decide what size paper to draw it on. When I worked at Chevrolet I took this to the max and prototyped entire vehicles years before the official prototype build phase.
When I transitioned to the world of networks and software I was shocked at how little prototyping there is. Website developers would deliver project plans that had two weeks of "QA" built in before the launch date. It was crazy, any major problem uncovered in that two weeks would delay the launch. It is still challenging to get a programmer to demonstrate a prototype early. So, I think you are on to something. Good luck with ARC!